Scholarlink

<p>Cognitive Radio Networks (CRNs) have the potential to significantly enhance spectrum utilisation by allowing unlicensed secondary users (SUs) to opportunistically access underused licensed bands. However, spectrum sensing under dynamic network conditions is still an issue with efficient and accurate communication. This paper proposes an Adaptive Deep Reinforcement Learning (ADRL) framework integrated with a Modified Snake Swarm optimisation Algorithm (MSSOA) to improve sensing efficiency, reward learning, and collision avoidance in CRNs. The ADRL model uses both Deep Q-Network (DQN) and Deep Recurrent Q-Network (DRQN) mechanisms. This lets each SU learn the best ways to choose channels and power-control strategies through experience replay and policy updates. Simulation results show that the proposed ADRL–MSSOA model performs remarkably better than conventional techniques such as DRQN-GWO, DBN and DNN on various parameters. When C₂ = 300, the reward parameter on the proposed model was 246.77 for DRQN-GWO, 223.85 for DBN, and 161.53 for DNN, which proved the improved stability of learning under a heavy load through the use of this optimal selection process. Even at reduced user variations (C₂ = 30 and C₂ = 3), the proposed framework maintained higher adaptability with rewards of 23.54 and − 8.65, respectively. In network performance evaluation, the model achieved a delivery ratio of 0.97, outperforming DRQN-GWO (0.6846), DBN (0.6373), and DNN (0.7855). Additionally, the model attains a delivery ratio of 0.97, a throughput improvement of 40%, and a delay reduction of 28% while minimising the collision rate below 5%. Overall, the proposed ADRL-MSSOA model the high adaptability faster convergence, and efficient spectrum sensing, making it a promising solution for intelligent self-learning 5G/6G cognitive communication systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive Deep Reinforcement Learning for Effective Spectrum Sensing in Cognitive Radio Networks

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • S. Kannan,
  • Suresh Arumugam,
  • P. Jyothilakshmi,
  • Sujai Paneerselvam,
  • V. Ravi Kumar,
  • P. Srujana,
  • Saroja M N

摘要

Cognitive Radio Networks (CRNs) have the potential to significantly enhance spectrum utilisation by allowing unlicensed secondary users (SUs) to opportunistically access underused licensed bands. However, spectrum sensing under dynamic network conditions is still an issue with efficient and accurate communication. This paper proposes an Adaptive Deep Reinforcement Learning (ADRL) framework integrated with a Modified Snake Swarm optimisation Algorithm (MSSOA) to improve sensing efficiency, reward learning, and collision avoidance in CRNs. The ADRL model uses both Deep Q-Network (DQN) and Deep Recurrent Q-Network (DRQN) mechanisms. This lets each SU learn the best ways to choose channels and power-control strategies through experience replay and policy updates. Simulation results show that the proposed ADRL–MSSOA model performs remarkably better than conventional techniques such as DRQN-GWO, DBN and DNN on various parameters. When C₂ = 300, the reward parameter on the proposed model was 246.77 for DRQN-GWO, 223.85 for DBN, and 161.53 for DNN, which proved the improved stability of learning under a heavy load through the use of this optimal selection process. Even at reduced user variations (C₂ = 30 and C₂ = 3), the proposed framework maintained higher adaptability with rewards of 23.54 and − 8.65, respectively. In network performance evaluation, the model achieved a delivery ratio of 0.97, outperforming DRQN-GWO (0.6846), DBN (0.6373), and DNN (0.7855). Additionally, the model attains a delivery ratio of 0.97, a throughput improvement of 40%, and a delay reduction of 28% while minimising the collision rate below 5%. Overall, the proposed ADRL-MSSOA model the high adaptability faster convergence, and efficient spectrum sensing, making it a promising solution for intelligent self-learning 5G/6G cognitive communication systems.